Motor bearing fault diagnosis based on CEEMDAN and GJO-optimized CNN

Abstract

To effectively extract fault features from motor bearings and improve the accuracy of fault diagnosis, this paper proposes a novel method for motor bearing fault diagnosis based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Golden Jackal Optimization (GJO) to optimize Convolutional Neural Networks (CNNs). First, given the highly non-stationary nature of motor bearing vibration signals, this paper proposes using CEEMDAN to decompose the bearing vibration signals. The vibration signals are decomposed into a series of IMF components. To effectively characterize the different operating states of the bearing, the fuzzy entropy of the calculated components is used as the fault feature vector. After performing dimension reduction and feature selection using Kernel Principal Component Analysis (KPCA), the feature vector is reconstructed. A CNN algorithm was adopted as the fault diagnosis model. To address the issue that the model’s main parameters significantly affect its classification performance, this paper employed GJO for parameter optimization and established a GJO-CNN model. Test set samples were then input into the model for fault diagnosis. Experimental results show that the proposed method achieved a comprehensive fault diagnosis accuracy of 99.64%. Compared with CNN, PSO-CNN optimized by the Particle Swarm Optimization (PSO) algorithm, the WOA-CNN model optimized by the Whale Optimization Algorithm (WOA), and the DBO-CNN model optimized by the Dung Beetle Optimization (DBO) algorithm, the proposed method achieves a 7.85%, 6.43%, 1.78%, and 2.5% improvement in overall fault diagnosis accuracy, respectively. This validates the effectiveness and superiority of the proposed method.